Summary
A bioRxiv preprint presents a machine-learning structural atlas of kinase inhibitor binding sites, built from 11,945 inhibitor complexes. The analysis suggests that evolutionary relationships between kinases can help predict shared access to allosteric drug pockets.
A bioRxiv preprint describes a machine-learning framework for mapping druggable binding pockets across the human kinome—the full set of protein kinases encoded by human genes. The study combines structural data from 11,945 kinase–inhibitor complexes with evolutionary relationships between kinases to identify where selective, non-ATP-competitive drugs may be developed.
The authors report that evolutionary proximity is a statistically significant predictor of whether kinases share access to allosteric pockets. Such pockets could provide alternatives to the highly conserved ATP-binding site, which is difficult to target selectively across the kinase family.
Why kinase selectivity is difficult
Protein kinases regulate cellular processes by transferring phosphate groups to other proteins. Because abnormal kinase activity is involved in cancer and other diseases, kinases are important therapeutic targets.
The challenge is that approximately 536 Manning kinase domains share a conserved ATP-binding pocket. A drug designed to occupy this common site can therefore interact with several kinases, increasing the difficulty of targeting one disease-related kinase while avoiding others.
Allosteric inhibitors approach the problem differently. Instead of competing with ATP at the main catalytic site, they bind to another pocket created by a particular kinase's shape and conformational state. These pockets can be more distinctive within individual kinases or narrower kinase subfamilies, offering a route to greater selectivity.
Classifying pockets with structural data
The researchers assembled a structural atlas from 11,945 kinase–inhibitor complexes. They trained an Extra Trees classifier—a machine-learning model based on an ensemble of decision trees—using how pocket residues interact with the bound ligand.
The model was used to assign all seven canonical kinase-binding modes and to distinguish subclasses of allosteric binding. This gives the analysis a common classification system for comparing pockets across the kinome rather than treating each inhibitor-bound structure as an isolated example.
The structural analysis found that about 303 Manning kinase domains had known inhibitors bound to them. That represents approximately 56.5% of the Manning kinase domain set. Among those 303 domains, 78 kinases—26%—were targeted by non-ATP-competitive allosteric inhibitors.
These figures describe kinase domains represented by inhibitor-bound structures in the study, rather than a count of approved medicines. The authors interpret the smaller allosteric fraction as evidence of substantial unexplored pharmacological space.
Using kinome evolution to find new targets
The study then linked the pocket classifications to the phylogeny of the human kinome. A phylogeny is an evolutionary relationship map: kinases that share a closer evolutionary history often retain related structural and functional features.
In the authors' analysis, evolutionary proximity was significantly associated with shared allosteric-pocket accessibility. The proposed principle is that a kinase with an experimentally observed or structurally characterised allosteric pocket may provide a useful guide to related kinases whose pockets have not yet been exploited.
The researchers call this approach “phylogenetic druggability transfer”. In practical terms, kinome evolution could become a prioritisation tool for selecting targets and binding sites before extensive medicinal-chemistry work begins. The framework is described as potentially useful not only for allosteric inhibitors, but also for orthosteric, covalent, bifunctional and chemical-degrader programmes.
From structural map to drug programmes
The work is a computational and structural preprint, so its immediate contribution is a classification and prediction framework rather than a tested medicine. The reported results can help identify candidate pockets and related kinases for follow-up experiments, while selective compounds and their therapeutic effects will require subsequent drug-discovery and biological validation.
The study therefore treats the kinome as more than a list of enzyme targets: it presents the evolutionary map as a way to organise opportunities for finding binding sites outside the conserved ATP pocket.